Intelligent Fault Diagnosis of Solar PV Systems Using Machine Learning Techniques

Abstract

Solar photovoltaic (PV) systems need effective fault detection system in order to sustain effective energy generation under different environmental and operating conditions. The proposed study is a data-driven study on fault categorization using machine learning methodologies of detecting faults in PV systems under nonlinear and noisy conditions. Electrical data of PV panels such as current voltage (I V ) and power voltage (P V ) of the panels were measured in three operating conditions: Healthy, Shading, and Open-Circuit. Important electrical characteristics that included open-circuit voltage, shortcircuit current, maximum power point voltage and current, fill factor, and some statistical indicators were determined to describe the behavior of the system.Naive Bayes and k-Nearest Neighbors (KNN) are two supervised learning algorithms that were used to identify faults. All of these models were optimized with the help of hyperparameter tuning and k-fold cross-validation to provide accurate performance. Comparative analysis revealed that the KNN classifier had greater detection capacity along with an accuracy of 95.47, whereas the Naive Bayes classifier had stable and consistent performance and an accuracy of 94.13, which proved to be robust in the nonlinearity and noise in measurements.On the whole, the findings suggest that Naïve Bayes and KNN models can be utilized in real-time detection of PV faults with a help of the easily measurable electrical parameters. The suggested solution is applicable in the practical monitoring of PV systems and it can endorses proactive maintenance measures, enhance reliability in the systems, and minimize energy waste.